Weighted Trigonometric Regression for Suboptimal Designs in Circadian Transcriptome Studies.

Area of Science:

• * Chronobiology and Genomics
• * Statistical Modeling in Biological Sciences

Background:

• * Circadian transcriptome studies commonly employ trigonometric regression to analyze gene expression patterns over a 24-hour cycle.
• * Optimal experimental design necessitates equispaced sampling, which is often logistically challenging in human studies.
• * Non-equispaced sampling can lead to variable statistical power and unreliable phase-shift parameter estimates.

Purpose of the Study:

• * To address the variability in statistical power caused by non-equispaced sampling in circadian transcriptome studies.
• * To introduce a weighted trigonometric regression method to mitigate power fluctuations.
• * To optimize the kernel density estimation hyperparameter using an E-optimality criterion.

Main Methods:

• * Development of a weighted trigonometric regression model.
• * Utilizing kernel density estimation to determine weights based on sample collection time density.
• * Implementing a search procedure for hyperparameter optimization guided by the E-optimality criterion.

Main Results:

• * Weighted regression demonstrated a consistent mitigation of power variability in hypothesis testing.
• * Simulations confirmed the effectiveness of the weighted approach.
• * Analysis of six circadian transcriptome datasets showed increased test statistics compared to unweighted regression (cosinor regression).

Conclusions:

• * Weighted trigonometric regression offers a robust solution for analyzing circadian gene expression data with non-equispaced samples.
• * The proposed method enhances the reliability and statistical power of analyses in chronobiology.
• * This approach is particularly beneficial for cosinor regression, a common tool in circadian research.